{"id":"W7155399693","doi":"10.13182/t133-49106","title":"HTR-PROTEUS: Interrogating an Old Benchmark with New (Modeling) Tricks","year":2025,"lang":"","type":"article","venue":"Transactions of the American Nuclear Society","topic":"Advanced Electron Microscopy Techniques and Applications","field":"Biochemistry, Genetics and Molecular Biology","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Alchemy (Canada)","funders":"","keywords":"Benchmark (surveying); Key (lock); History of computing","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0001147287,0.0002942803,0.0003403411,0.00003205143,0.0005412801,0.00003864419,0.0007029864,0.0001388021,0.00004561453],"category_scores_gemma":[0.000007433231,0.0002479723,0.0004868544,0.0007991461,0.0006967185,0.00001805172,0.00005746252,0.0004846248,0.000001148521],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008995603,"about_ca_system_score_gemma":0.000326313,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008262044,"about_ca_topic_score_gemma":0.0001396701,"domain_scores_codex":[0.9984911,0.0000576455,0.0003914645,0.0005460286,0.0001416618,0.0003721148],"domain_scores_gemma":[0.9983714,0.00001310305,0.000345315,0.001033137,0.0001339414,0.0001030639],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.0004147624,0.0005340629,0.0001298116,0.00009364718,0.0006956811,1.07124e-7,0.001694851,0.01208233,0.8791773,0.0002793974,0.001965939,0.1029321],"study_design_scores_gemma":[0.003391117,0.006488915,0.0002891979,0.001038877,0.001939033,0.00004271091,0.0344205,0.09703939,0.7483363,0.001634681,0.1031692,0.002210017],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3725834,0.000316012,0.6246172,0.001431735,0.00004531471,0.0005306522,0.00004115617,0.00004587302,0.0003886269],"genre_scores_gemma":[0.9377432,0.001396268,0.05853528,0.0009165754,0.00005197562,0.000028132,0.00001056868,0.00005313699,0.001264813],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5660819,"threshold_uncertainty_score":0.9999973,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.005716994262216237,"score_gpt":0.291335034737866,"score_spread":0.2856180404756498,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}